COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	AI: THE SECOND WAVE
C00015 ENDMK
C⊗;
AI: THE SECOND WAVE

Proposed questions for 2-part interview with Dr. John McCarthy

Part I

1. Dr. McCarthy, you're regarded as the father of AI, so this question is
especially appropriate. There is a wide range of definitions for artificial
intelligence used in the field--many of them contradictory. How would you
definite ``artificial intelligence'' and how does this relate to what we mean by
human intelligence?

2. Alan Turing began his 1950 article ``Computing Machinery and intelligence''
with the question, ``Can machines thing?'' In what sense of the word do computers
``think?'' Does the computer function in some ways like a human brain?

3. Artificial Intelligence--although new as a science--is obviously making enormous
strides. Can you tell us why for some people the subject of artificial intelligence
is still controversial?

4. Would you describe for us the various components of AI?

5. What is the distinction between research AI and applied AI?

6. The historic 1956 Dartmouth Conference--which you organized--is looked on as
the place where the AI revolution was born. What were some of the obstacles that
AI had to overcome to be accepted in the forefront of scientific research?

7. What is the distinction between AI as a science and computer science?

8. In 1958 you invented  what went on to become the most widely used symbolic
processing language in the world--LISP. What made it different from other
programming languages? Would you describe how LISP relates to the history of AI?

9. Since the time of the 1956 Conference, what in your opinion is the most 
exciting direction that AI research has taken?

10. What are the possibilities for technologies based on the present state of
AI science?

Part II

1. One of the AI controversies is whether or not it is possible to create a machine
that can use ``common sense'' reasoning. Do you feel such a development is
possible? From a research standpoint, what are the most difficult obstacles to
be overcome in the area of creating machines that use common sense knowledge and 
reasoning?

2. We've examined the various applied disciplines within AI: voice recognition,
natural language, robotics, and expert systems.  For each of these disciplines, 
what do you see as the most exciting developments on the horizon today?

3. Other countries, notably Britain and Japan, are making great strides in AI 
research. Would you describe some of these international developments?

4. What are you working on currently?

5. What do you perceive as the greatest challenge to the development of AI today?

6. What do you forsee from AI in the next five to ten years--and in the next fifty
years--in terms of impact on how we do business and way we live?

JMC

1. This is one of these cases where paternity is uncertain. Turing, Newell, and
Minsky are also paying child support. 

There really isn't just one father. After WWII, quite a few people started work.
I organized the first conference, coined the name and originated the approach 
based on mathematical logic.

Artificial intelligence as science is concerned with methods for the achievement
of goals in complex situations of partial information and partial by defined 
concepts.  Humans often solve such problems and the methods that have to be
used because of the nature of the problem are often the same for man and
machine.

AI as technology apples what is now known about intelligence to making useful
computer programs for tasks that have a kind of complexity for which standard
computer programming isn't very effective.

2. Both people and computers decide what to do using the facts they know, but 
computers are still rather feable minded.

We presently understand SOME of the intellectual mechanisms well enough to put them
in computer programs.  You can call what these programs do thinking or you can
hold out for more--not agree to call it thinking till till machines do every
intellectual activity humans do.

Many computer programs processes are like some human mental processes although 
there  are enormous but mostly unknown differences at the mental level.

3. Some philosophical views find AI impossible in principle. 
Some find exaggerated  claims for the present state of the art.
Some object to machines being used to make judgments.
Some fear the robot will take over.
Some object to AI as part of a general objection to technology and even to American
society.

4. Some of the more prominent are:
pattern recognition--visual, aural and situational
reasoning--drawing consequences from beliefs
planning--deciding actions that will achieve goals
learning--from experience and from communication

5. The same as between science and engineering in other domains. The one emphasizes
new understanding--the other what useful things can be done with present 
understanding.

6. We had  to make some scientific progress, but also some of us had to develop white
hair.

7. AI is part of computer science--the part where the computer isn't given a model
of the phenomenon in advance but must find it for itself.

8.  LISP was designed for computing with the complex symbolic structures needed to
express thoughts rather than with numbers. Therefore, its basis operations are 
different. Now LISP and its variants have a rival--namely, Prolog.

9. I still like the use of mathematical logic to express facts and reasoning.
Developing a precise idea of non-monotonic reasoning in which conclusions are
drawn from the absence of certain information is the single most important
logical idea AI has contributed. I believe it will affect the way we think about
philosophy and about human decision making.

10. The best tecchnological possibility for AI right now is decision making based
on large numbers of facts used in a relatively shallow way.  An example is 
deciding whether to allow a credit card charge based on the risks both of
fraud and nonpayment and also suggesting ways of settling doubts.  Inference Corp.
is doing such a system for American Express.

II

1. Human level

Yes, computer common sense is possible but not easy. The biggest difficulty seems
to be making the computers deal with so called open systems, where the concepts
are not given precisely in advance.

2. That's too many to talk about them all. I see natural language as having many
features we need to put in our formal languages such as certain kinds of dependence 
on context.

3. Logic programming started in Britain and France and is now emphasized by a major
Japanese project. The British have done some good work in common sense reasoning
and the Japanese are starting.  Both also do good work in robotics. The US is
still ahead though, in a large part because of the support of the Defense 
Department.

4. I am working on non-monotonic reasoning and making a logical treatment of
contexts.

5. Programs that can extend the context of their computations in response to new
information. 

6. There are conceptual problems not even formulated. AI is a very hard problem.
When people ask me how long before computers programs will have human level 
intelligence, I can only say between 5 and 500 years. It was 100 years from
Mendel to the genetic code. I think a number of ``heuristic classification''
programs will become economically important in 5 to 10 years. So will 
industrial robots using vision. I doubt that we'll get household servant robots
before 50 years. I think a personal advisor will shortly be feasible--for example,
a program that will suggest achieving several goals with one trip and will reduce
the chance of blunders.

I think the defense oriented projects supported by DARPA have a good chance of 
producing useful results.

A system for commercial communication among computers belonging to different 
companies will need a surprising amount of AI, especially non-monotonic
reasoning, but it can be developed.